MediumVulnerabilityMonitoring

SANS ISC reports scans targeting LiteLLM and Starlette CVEs

SANS ISC reports honeypot requests that appear to target LiteLLM and Starlette CVEs; check Starlette is at 1.0.1 or later and review exposure.

Published
Fixed in
Starlette 1.0.1 or later per SANS ISC (LiteLLM fixed version not confirmed)
Exploitation
Yes, attempts only. SANS ISC says its honeypots saw requests that appear to try exploitation; success against real systems is not confirmed.

Live scores

CVE-2026-42271
EPSS
0.914
very likely to be attacked soon
VEDAS
0.975
strong exploit evidence
On CISA KEV
Yes, since 2026-06-08
Detection rule
Nuclei · Suricata
EPSS (chance of attack)
100.1CVE-2026-42271 EPSS over time2026-082026-11-17
VEDAS (exploit evidence)
100.5CVE-2026-42271 VEDAS over time2026-082026-11-17

Thin horizontal line: the “likely exploited” threshold (EPSS 0.10, VEDAS 0.5). Shaded “no data” areas are periods when cve-scores did not publish.

Live scores

CVE-2026-48710
EPSS
0.703
very likely to be attacked soon
VEDAS
0.980
strong exploit evidence
On CISA KEV
Yes, since 2026-09-02
Detection rule
Nuclei · Suricata
EPSS (chance of attack)
100.1Added to CISA KEV 2026-09-02CVE-2026-48710 EPSS over time2026-082026-11-17
VEDAS (exploit evidence)
100.5Added to CISA KEV 2026-09-02CVE-2026-48710 VEDAS over time2026-082026-11-17

Thin horizontal line: the “likely exploited” threshold (EPSS 0.10, VEDAS 0.5). Shaded “no data” areas are periods when cve-scores did not publish. The vertical line is the day CISA added it to its exploited list.

VEDAS flagged this bug about 25 days before CISA confirmed attacks.

What happened

According to the SANS Internet Storm Center, its honeypots began receiving requests for /mcp-rest/test/tools/list over the weekend before the 2026-11-16 diary. Handler Johannes Ullrich says these requests “appear to attempt to exploit” CVE-2026-42271 and CVE-2026-48710. The diary says Horizon3.ai published a short write-up on chaining these two vulnerabilities for unauthenticated remote code execution. That write-up was not among the sources reviewed here, so its content is only as SANS ISC describes it.

SANS ISC reports the following about the observed traffic:

  • The requests are POSTs with a placeholder Bearer token and a JSON body that names a “stdio” transport and runs the command “id”.
  • The Host headers are unusual: localhost], [localhost, 127.0.0.1] and litellm.proxy.
  • The diary suggests the bracket characters may be an attempt to evade filters built around the public proof-of-concept exploit.

On the cause, SANS ISC says Starlette versions prior to 1.0.1 do not validate the Host header (the issue is known as “BadHost”). Starlette uses the unvalidated value to build a full URL, which is then used for access control decisions. The public PoC reportedly uses characters such as /, ? and # in the Host header to confuse that logic. SANS ISC says that for LiteLLM this exposes the /mcp-rest/test/tools/list endpoint, which normally requires authentication.

Not confirmed: whether any of the observed requests succeeded against a real system, who is behind them, how many hosts are targeted, and which LiteLLM versions are affected or fixed. SANS ISC also notes this is not a new vulnerability and that exploitation was reported back in July.

Am I affected?

Based on the SANS ISC diary, you may be affected if:

  • You run LiteLLM, or another application that relies on Starlette for access control, and
  • Your Starlette version is earlier than 1.0.1.

The diary does not list affected LiteLLM versions or a LiteLLM fix, so those are not confirmed. A LiteLLM proxy reachable from the internet is the situation the observed scans appear to target, though the diary does not state this outright. SANS ISC does not say whether the scans have hit any production systems.

What to do now

  1. Check which Starlette version your LiteLLM deployment, or any other Starlette-based service, uses. SANS ISC says versions before 1.0.1 do not validate the Host header.
  2. Upgrade Starlette to 1.0.1 or later where you can, and check the vendor advisories for LiteLLM, since no LiteLLM fixed version is given in the sources.
  3. Limit exposure of the proxy and its /mcp-rest/ paths to trusted networks, in line with SANS ISC’s advice to reduce attack surface.
  4. Review web and proxy logs for POST requests to /mcp-rest/test/tools/list and for Host headers with unusual characters, such as the values listed above.
  5. If you maintain WAF rules, follow the SANS ISC suggestion: flag characters that should never appear in a hostname, or allow only specific hostnames, rather than matching one exploit’s pattern.

Sources